Tokenization of text data to facilitate automated discovery of speech disfluencies

Abstract

Introduced here are computer programs and associated computer-implemented techniques for discovering the presence of filler words through tokenization of a transcript derived from audio content. When audio content is obtained by a media production platform, the audio content can be converted into text content as part of a speech-to-text operation. The text content can then be tokenized and labeled using a Natural Language Processing (NLP) library. Tokenizing/labeling may be performed in accordance with a series of rules associated with filler words. At a high level, these rules may examine the text content (and associated tokens/labels) to determine whether patterns, relationships, verbatim, and context indicate that a term is a filler word. Any filler words that are discovered in the text content can be identified as such so that appropriate action(s) can be taken.

What is claimed is: 
     
         1 . A non-transitory computer-readable medium with instructions stored thereon that, when executed by a processor of a computing device, cause the computing device to perform operations comprising:
 receiving, as input, audio data that includes words uttered by an individual;   obtaining text data that is representative of a transcript created for the audio data;   tokenizing each word in the text data to produce a set of tokens that are arranged in sequential order;   performing grammatical tagging so that each token in the set of tokens is labeled as corresponding to a part of speech; and   applying a rule associated with a filler word that is representative of a non-lexical utterance to the set of tokens,
 wherein the rule is representative of a data structure that specifies (i) the filler word and (ii) a contextual parameter indicative of a criterion that must be satisfied for the rule to indicate that a given token represents an instance of the filler word. 
   
     
     
         2 . The non-transitory computer-readable medium of  claim 1 , wherein the text data is produced by performing a speech-to-text operation on the audio data. 
     
     
         3 . The non-transitory computer-readable medium of  claim 1 , wherein the text data is acquired from a source external to the computing device. 
     
     
         4 . The non-transitory computer-readable medium of  claim 3 , wherein the text data is acquired from a database managed by a transcription service in response to a request for the transcript of the audio data. 
     
     
         5 . The non-transitory computer-readable medium of  claim 1 , wherein the text data includes at least one of: reviews, surveys, evaluations, posts on social media platforms, books, or news articles. 
     
     
         6 . The non-transitory computer-readable medium of  claim 1 , wherein the text data is tokenized using a Natural Language Processing (NLP) library. 
     
     
         7 . The non-transitory computer-readable medium of  claim 1 , wherein an appropriate tokenization scheme for said tokenizing is determined based on a language of the text data. 
     
     
         8 . The non-transitory computer-readable medium of  claim 1 , wherein the operations further comprise:
 performing language identification based on one or more classifiers that use short-character subsequences as features; and   determining, based on an outcome of said performing, that the text data is in a default language for which tokenization is possible.   
     
     
         9 . The non-transitory computer-readable medium of  claim 1 , wherein the operations further comprise:
 identifying, based on an outcome of said applying, a word in the text data that is representative of an instance of the filler word.   
     
     
         10 . A non-transitory computer-readable medium with instructions stored thereon that, when executed by a processor of a computing device, cause the computing device to perform operations comprising:
 tokenizing each word in text data to produce a set of tokens that are arranged in sequential order;   performing dependency parsing so that for each sentence included in the text data, a dependency parse is extracted and a grammatical structure is defined;   for each token in the set of tokens,
 labeling the token based on the dependency parse and the grammatical structure for the sentence with which the token is associated; 
 applying a rule associated with a filler word that is representative of a non-lexical utterance to the set of tokens,
 wherein the rule is representative of a data structure that specifies (i) the filler word and (ii) a contextual parameter indicative of a criterion that must be satisfied for the rule to indicate that a given token represents an instance of the filler word; and 
 
   executing an action for each instance of the filler word identified by the rule.   
     
     
         11 . The non-transitory computer-readable medium of  claim 10 , wherein said executing comprises:
 causing display of the text data on an interface in such a manner that each word that represents an instance of the filler word is visually distinguishable from other words in the text data.   
     
     
         12 . The non-transitory computer-readable medium of  claim 10 , wherein said executing comprises:
 removing, from the text data, each instance of the filler word identified by the rule.   
     
     
         13 . The non-transitory computer-readable medium of  claim 12 , wherein said executing further comprises:
 identifying an audio file that is associated with the text data; and   for each instance of the filler word identified by the rule,
 removing, from the audio file, a segment that corresponds to that instance of the filler word. 
   
     
     
         14 . The non-transitory computer-readable medium of  claim 13 , wherein the audio file and the text data are dynamically linked such that a change in the text data is automatically reflected on the audio file, and vice versa. 
     
     
         15 . A method for distinguishing valid repetitions of a phrase from invalid repetitions of the phrase in audio data, the method comprising:
 obtaining a set of tokens that are arranged in sequential order,
 wherein each token is representative of a corresponding word in a transcript that is representative of a transcript created for the audio data; and 
   applying, to the set of tokens, a rule associated with the phrase so as to obtain a set of outputs, each of which indicates whether a corresponding token in the set of tokens is representative of an invalid repetition.   
     
     
         16 . The method of  claim 15 , wherein the rule is representative of a data structure that includes:
 (i) the phrase,   (ii) a first criterion that specifies a first occurrence of the phrase must be immediately followed by a second occurrence of the phrase for the rule to indicate that a repetition exists, and   (iii) a second criterion that specifies a certain form of punctuation must occur before the first occurrence of the phrase, after the second occurrence of the phrase, or between the first and second occurrences of the phrase to distinguish valid repetitions from invalid repetitions.   
     
     
         17 . The method of  claim 15 , further comprising:
 performing a remediation action for each invalid repetition of the phrase in the set of tokens.   
     
     
         18 . The method of  claim 17 , wherein the remediation action comprises deleting each token that is representative of an invalid repetition in the set of tokens. 
     
     
         19 . The method of  claim 15 , further comprising:
 updating the rule to ignore valid repetitions of the phrase, if any.   
     
     
         20 . The method of  claim 15 , wherein the rule is a first rule in a set of rules and the set of outputs is a first set of outputs of multiple sets of outputs, the method further comprising:
 applying, to the set of tokens, a second rule in the set of rules so as to obtain a second set of outputs, each of which indicates whether a corresponding token in the set of tokens is representative of an invalid repetition of another phrase that is different than the phrase.